Over the past decade, the world has seen an explosion of online and social activity, all of which is generating vast amounts of data. Louis Rossouw surveys the future for insurance.
- Predictive modelling. This concerns the process of predicting future probabilities (e.g, death, health events or policies lapsing) based on data. This technique is well known to insurance companies and is being integrated into insurance operations pricing mortality and other risks as well as modelling persistency, cross-selling opportunities and similar variables. It is an especially powerful tool when linking-in additional data about the consumer (either internal data or external data, such as health behaviour or data linked to location information).
- Data clustering. This technique allows for the automated grouping of similar data points. A good visual example of this type of analysis is livehoods.org, using both geographic distance and the similarity of people visiting places within an area to ascertain true neighbourhood boundaries in an automated fashion.
Clustering techniques are also used by music applications to automatically classify your music into various genres based on these techniques (for example, the Genius Mix feature of iTunes).
In the insurance sphere, this technique can be applied to underwriting data to cluster cases with impairments and therefore gain more understanding of the more common multiple impairments seen in similar cases. These techniques are also often used in the customer segmentation process.
- Data integration. In this process, additional data is linked into existing analysis to improve predictive power of modelling. An example is the use of banking data as additional predictive variable in a persistency model.
- A/B testing. This technique emerged from companies attempting to optimise their websites by removing the guesswork. Given a choice between design or content items to have on a page, the company could guess which is most appropriate to its visitors. The alternative is to actually test both versions of the website on a sample of visitors and observe which produces the best results (typically measured by clicks through to the content). This data is then used to decide which design to use or content to run. A life insurer, for example, could measure changes in process and underwriting before implementing them.
- Sentiment analysis. Using textual keyword analysis and other techniques, researchers are able to analyse the mood or sentiment of Twitter status updates on a given topic, geography or time period. The status of a brand (or brands) can also be measured in this way.
- Data visualisation. Given the amounts of data that are becoming available and the increased graphic powers of most computers, many interesting visualisations of data are available, which allow for more intuitive understanding of data and relationships.
